Fordel StudiosThis case study describes a real engagement. Client identity, proprietary details, and specific metrics are anonymized or approximated under NDA.
AI-Enhanced Property Discovery Platform
High bounce rate on property listings combined with a 60% no-show rate for in-person visits. The platform had strong organic traffic but thin conversion: buyers had no way to evaluate properties in depth before committing to a visit, and agents were following up cold with no context about buyer intent.
Virtual tour system with smart property highlighting, automated room-by-room descriptions, and integrated lead qualification that captures buyer preference signals throughout the tour and delivers a contextualized handoff to agents.
This engagement addressed the gap between listing interest and qualified agent engagement by building a property exploration experience that captures intent signals throughout the viewing session. The system was built in three layers: an asset processing pipeline that converts raw 360-degree imagery into optimized tour assets with AI-generated room classifications and feature tags, a Three.js viewer with adaptive performance for mid-range mobile devices, and a progressive lead qualification flow embedded in the tour experience. The system was designed to process the platform's existing 5,000+ listings without per-property manual work — all content generation is automated from the asset ingestion pipeline.
The Challenge
The primary engineering constraint was mobile performance. The platform's user base skews heavily toward mid-range Android devices (3GB RAM), and full-quality cubemap WebGL rendering caused frame drops and thermal throttling on these devices. Solving this without degrading the desktop experience required adaptive quality selection at load time based on device capability detection. The lead qualification flow presented a UX challenge: capturing meaningful intent signals without interrupting the tour experience required several rounds of iteration to get the friction-to-signal ratio right. At 5,000+ active listings, content generation had to be fully automated — any per-listing manual step would have made the system operationally unviable at scale.
How We Built It
Asset audit and ingestion pipeline (Weeks 1–3): The platform had 360-degree images for approximately 60% of listings but had never used them. We built an ingestion pipeline that converts raw equirectangular images into optimized cubemap format, runs an OpenAI Vision pass to classify each room (living room, bedroom, kitchen, bathroom, balcony) and extract feature tags (natural light, storage quality, finishes, views), and stores processed assets and metadata in object storage. For the remaining 40% of listings without 360-degree assets, the pipeline generates an enhanced photo gallery experience with AI-generated room descriptions from flat images — a lower-fidelity fallback that still outperforms the original photo carousel.
Three.js viewer and mobile optimization (Weeks 4–7): We built a custom Three.js WebGL viewer embedded in the Next.js listing page. The viewer renders cubemap panoramas with hotspot overlays for room-to-room navigation and surfaces AI-generated feature callouts when the viewer dwells on specific areas. Adaptive quality selection detects device capability on load and serves appropriately compressed assets. Lazy loading defers rooms not yet navigated to, and a canvas-size cap applies on low-memory devices. The result was consistent 30fps rendering on 3GB RAM devices, covering the majority of the Android user base.
Automated content generation (Weeks 8–10): For each listing, the pipeline generates three assets: a room-by-room narrative description grounded strictly in detected features (no hallucinated details), a bullet-point highlights list for the listing card, and a set of standardized comparison attributes (floor area, room count, parking, amenities) extracted from the unstructured seller-submitted listing data. These assets are generated on ingestion, stored in PostgreSQL, and served via the existing listing API with no changes to the frontend data model beyond new fields.
Lead qualification flow and agent handoff (Weeks 11–14): A progressive qualification flow surfaces naturally within the tour: a room-save interaction that captures preference signals, a budget confirmation after 3+ minutes of active viewing, and a scheduling prompt after tour completion. The agent handoff delivers a context packet before first contact: rooms the buyer spent most time in, features they saved, stated budget, and preferred visit window. This replaces cold outreach with an informed conversation. Visit scheduling volume dropped but no-show rate fell from 60% to approximately 29%.
What We Delivered
Average time on page for tour-enabled listings increased 2.1x. Bounce rate on those listings fell by 34%. The proportion of submitted leads classified by agents as qualified on first contact increased from 31% to 43%, reflecting the value of the pre-call context packet. Agents consistently reported arriving at calls with specific questions already answered rather than needing a full discovery conversation.
In-person visit dynamics shifted. Scheduled visit volume dropped, but no-show rate fell from 60% to 28%, resulting in a net increase in productive visits per agent per week. Buyers who completed a virtual tour arrived with formed intent rather than general curiosity, making in-person visits shorter and more decision-focused.
The asset ingestion pipeline processes new listings automatically at submission, meaning the content generation overhead scales at zero marginal cost. The Three.js viewer has been extended to handle video walkthroughs (submitted alongside 360-degree assets for newer listings) using the same adaptive delivery infrastructure built for the cubemap renderer.
Ready to build something like this?
Tell us what you are building. We will scope it, price it honestly, and give you a clear plan.
Start a ConversationFree 30-minute scoping call. No obligation.